Explore the Phytophthora trait database
Louise Barwell
22 May 2019
This document is intended to help users reproduce the analyses in Barwell et al. (2019) A conceptual trait-based framework for invasion risk assessment of oomycete plant pathogens (in prep.). We explore the Phytophthora trait database and it’s potential uses for plant health risk assessment. Using an ITS phylogeny, We fit models of trait evolution, visualise the phylogenetic distribution of traits and quantify the evolutionary lability of each trait. We then explore patterns of trait covariance and resource allocation trade-offs in the context of biotic niche partitioning, abiotic environmental filtering and movement of species.
Trait data and phylogeny
The trait data is stored in the file ‘2018-01-29_Phytophthora_trait_database_merged.csv’.
Continuous traits
## [1] "oospore_wall_thickness" "oospore_wall_index"
## [3] "gr_at_opt" "temp_min"
## [5] "temp_opt" "temp_max"
## [7] "temp_range"Discrete traits
## [1] "papillate" "proliferation" "sporangiophore"
## [4] "caduceus" "pedicel_length" "chlamydospores"
## [7] "hyphal_swelling" "oospores" "ho_he_s"
## [10] "oogonia_wall" "tapering_base" "amphi_para"
## [13] "aplerotic_plerotic" "oospore_wall_cat" "gr_at_opt_cat"
## [16] "temp_profile" "temp_range_cat"The phylogeny is stored in the file ‘updated - Posterior output.newick’. It was created by Treena Burgess on the 28/01/2018 and now includes the new species described in 2017 and 2018.
##
## Phylogenetic tree with 179 tips and 147 internal nodes.
##
## Tip labels:
## P. acerina, P. caryae, P. plurivora, P. pini, P. pachypleura, P. capensis, ...
##
## Rooted; includes branch lengths.Is it rooted?
## [1] TRUEIt is bifurcating?
## [1] FALSEThere are some polytomies. The functions to fit trait evolution models requirebifurcating trees. Clade 9 and 10 are not well-resolved. To resolve them, first identify the most recent commone ancestor of clades 9 and 10. There are three ways to resolve this node. We will use the one that correctly resolves species into their recognised clades.
Is it ultrametric?
## [1] FALSENo, we’ll Estimate the ages of the nodes by assuming branch lengths are proportional to the number of substitutions.
How many species are present in both the trait database and the phylogeny?
## [1] 179## [1] "P. cactorum" "P. clandestina"
## [3] "P. hedraiandra" "P. idaei"
## [5] "P. infestans" "P. ipomoeae"
## [7] "P. iranica" "P. mirabilis"
## [9] "P. nicotianae" "P. phaseoli"
## [11] "P. pseudotsugae" "P. tentaculata"
## [13] "P. x andina" "P. x pelgrandis"
## [15] "P. x serendipita" "P. 'acacia'"
## [17] "P. 'aquatilis'" "P. 'emzansi'"
## [19] "P. 'mekongensis'" "P. acerina"
## [21] "P. amaranthi" "P. bisheria"
## [23] "P. botryosa" "P. capensis"
## [25] "P. capsici" "P. caryae"
## [27] "P. citricola" "P. citrophthora"
## [29] "P. colocasiae" "P. elongata"
## [31] "P. frigida" "P. glovera"
## [33] "P. himalsilva" "P. meadii"
## [35] "P. mengei" "P. mexicana"
## [37] "P. multivesiculata" "P. multivora"
## [39] "P. occultans" "P. pachypleura"
## [41] "P. pini" "P. plurivora"
## [43] "P. siskiyouensis" "P. terminalis"
## [45] "P. tropicalis" "P. ilicis"
## [47] "P. nemorosa" "P. pluvialis"
## [49] "P. pseudosyringae" "P. psychrophila"
## [51] "P. alticola" "P. arenaria"
## [53] "P. boodjera" "P. litchii"
## [55] "P. megakarya" "P. palmivora"
## [57] "P. quercetorum" "P. agathidicida"
## [59] "P. castaneae" "P. cocois"
## [61] "P. heveae" "P. 'paludosa'"
## [63] "P. 'personii'" "P. 'raspberry'"
## [65] "P. 'walnut'" "P. amnicola"
## [67] "P. asparagi" "P. balyanboodja"
## [69] "P. bilorbang" "P. borealis"
## [71] "P. chlamydospora" "P. condilina"
## [73] "P. cooljarloo" "P. crassamura"
## [75] "P. fluvialis" "P. gemini"
## [77] "P. gibbosa" "P. gonapodyides"
## [79] "P. gregata" "P. humicola"
## [81] "P. inundata" "P. kwongonina"
## [83] "P. lacustris" "P. litoralis"
## [85] "P. megasperma" "P. mississippiae"
## [87] "P. moyootj" "P. ornamentata"
## [89] "P. pinifolia" "P. pseudorosacearum"
## [91] "P. riparia" "P. rosacearum"
## [93] "P. thermophila" "P. x stagnum"
## [95] "P. asiatica" "P. attenuata"
## [97] "P. cajani" "P. cambivora"
## [99] "P. cinnamomi" "P. europaea"
## [101] "P. flexuosa" "P. formosa"
## [103] "P. fragariae" "P. fragariaefolia"
## [105] "P. intricata" "P. melonis"
## [107] "P. nagaii" "P. niederhauserii"
## [109] "P. parvispora" "P. pisi"
## [111] "P. pistaciae" "P. rubi"
## [113] "P. sojae" "P. tyrrhenica"
## [115] "P. uliginosa" "P. uniformis"
## [117] "P. vignae" "P. vulcanica"
## [119] "P. x alni" "P. x heterohybrida"
## [121] "P. x incrassata" "P. x multiformis"
## [123] "P. 'castitis'" "P. 'kelmania'"
## [125] "P. 'parsley'" "P. austrocedri"
## [127] "P. brassicae" "P. cichorii"
## [129] "P. cryptogea" "P. dauci"
## [131] "P. drechsleri" "P. erythroseptica"
## [133] "P. foliorum" "P. hibernalis"
## [135] "P. lactucae" "P. lateralis"
## [137] "P. medicaginis" "P. obscura"
## [139] "P. porri" "P. primulae"
## [141] "P. pseudocryptogea" "P. pseudolactucae"
## [143] "P. ramorum" "P. sansomeana"
## [145] "P. syringae" "P. trifolii"
## [147] "P. 'prodigiosa'" "P. aquimorbida"
## [149] "P. captiosa" "P. chrysanthemi"
## [151] "P. constricta" "P. estuarina"
## [153] "P. fallax" "P. hydrogena"
## [155] "P. hydropathica" "P. insolita"
## [157] "P. irrigata" "P. macilentosa"
## [159] "P. macrochlamydospora" "P. parsiana"
## [161] "P. polonica" "P. pseudopolonica"
## [163] "P. quininea" "P. rhizophorae"
## [165] "P. richardiae" "P. virginiana"
## [167] "P. boehmeriae" "P. gallica"
## [169] "P. gondwanensis" "P. intercalaris"
## [171] "P. kernoviae" "P. morindae"
## [173] "P. 'ohioensis'" "P. castanetorum"
## [175] "P. quercina" "P. tubulina"
## [177] "P. versiformis" "P. lilii"
## [179] "P. stricta"How many missing values for each trait?
## species_name phylo_clade phylo_sub_clade
## 0 0 0
## date papillate proliferation
## 5 0 0
## sporangiophore caduceus pedicel_length
## 0 0 138
## chlamydospores hyphal_swelling oospores
## 0 0 0
## ho_he_s oogonia_wall tapering_base
## 0 20 21
## amphi_para aplerotic_plerotic oospore_wall_thickness
## 21 21 22
## oospore_wall_index oospore_wall_cat gr_at_opt
## 22 22 30
## gr_at_opt_cat temp_min temp_opt
## 30 3 1
## temp_max temp_profile temp_range
## 1 1 3
## temp_range_cat
## 3How many are known hybrids?
## [1] 8## [1] "P. x pelgrandis" "P. x serendipita" "P. x andina"
## [4] "P. x heterohybrida" "P. x multiformis" "P. x alni"
## [7] "P. x incrassata" "P. x stagnum"Modelling trait evolution
Univariate comparison of evolutionary models
Below, we fit nine models of trait evolution to each of the 24 ecological and morphological traits to test which evolutionary theory best predicts the observed distribution of trait values. The likelihood of each model fit is compared using AIC framework, so that models with additional parameters are penalised and must signifantly improve model fit to justify their inclusion. The figures below show \(\Delta AIC\) values for each of the models. A value of \(\Delta AIC = 0\) represents the best model for that trait.
Fig. 1 Comparison of trait evolution models for seven continuous traits of 177 Phytophthora species. The nine models of trait evolution are compared using AIC with a correction for small sample sizes. Models are ranked based on \(\Delta AIC\), with lower values indicating better model fit to the observed distribution of trait values. Interpreting the parameters \(\lambda\), \(\kappa\) and \(\delta\) as described in the section “Interpreting trait evolution models” is valid only if those models perform well relative to other models. Among the continuous traits, the lambda model generally performs the best of all the models. Low values of \(\lambda\) can be interpreted as evidence for selection pressures, which disrupt the pattern of trait covariance expected if trait divergence was proportional only to the amount of shared phylogenetic history (Brownian Motion). Among the continuous traits \(\lambda\) is usually quite close to 1, indicating the observed trait values are strongly structured by the amount of shared phylogenetic history between two species. The kappa transformation also performs relatively well. Many of the values are closer to 0 which could be interpreted as evidence for short periods of rapid evolution followed by longer periods of stasis in which no change occurs (punctuated equilibrium). The values of \(\delta\) are generally much higher than 1, which could indicate that recent evolution in these traits has been especially rapid. However, the delta model doesn’t generally perform that well, so inferring anything about recent rates of evolution relative to more basal part of the tree might be invalid. The colours of the bars correspond to the models described in the section “Interpreting trait evolution models”.
Fig. 2 Comparison of trait evolution models for 17 discrete traits of 177 Phytophthora species. Note the y axis is log \(1+ \Delta AIC\) The six models of trait evolution are compared using AIC with a correction for small sample sizes. Models are ranked based on \(\Delta AIC\), with \(\Delta AIC = 0\) indicating the best fitting model to the observed distribution of trait values. In general, either the kappa or lambda transformations of the tree seem to best fit the distribution of the categorical traits across the tree. \(\lambda\) is generally rather high across the categorical traits, suggesting there is a strong phylogenetic signal in most traits. For kappa, values are generally low, which is evidence that traits may evolve in rapid bursts, punctuated by long periods of stability. There may be some issues with convergence of lambda that we need to address before we can compare models of trait evolution for categorical traits with confidence (e.g. parameter estimeates for proliferation trait are at the limits of the optimisation search bounds). The method is generally applied to continuous traits. When applied to categorical traits, we have to make assumptions about the rates of transition from one state to another and whether this varies between traits or is more likely in one direction than another. Tweaking this part of the model may help address the convergence issues. The only version I had time to run was an “equal-rates” model with a single parameter for all transition rates. The colours of the bars correspond to the models described in the section “Interpreting trait evolution models”.
Summarise the number of times each model is within 5AIcc units of the best-supported model.
The \(\lambda\) branch length transformation model of trait evolution consistently performs well across all traits and seems, overall, to be the best model. \(\kappa\) and \(\delta\) transformations also perform resoanbly well across traits. However, given that \(\kappa\) and \(\delta\) parameters are likely to be biased when taxa are not fully sampled (e.g. when there are extinct or undescribed species missing from the phylogeny), and the \(\lambda\) parameter is not, we’ll base our phylogenetic signal inferences on \(\lambda\).
##
## lambda kappa EB delta BM white OU
## 19 13 9 8 6 3 1Interpreting trait evolution models
In theory, comparison of these models should allow us to link patterns in the phylogeny with evolutionary processes (Pagel 1999 Nature 401:877-884), but there do seem to be quite a few pitfalls in interpreting the model parameters and what these mean in terms of the underlying evolutionary processes (e.g. see Cooper et al. 2016 Biol. J. Linn. Soc. 118: 64-77)
The models make the following assumptions about the evolutionary processes generating variation in traits.
Brownian Motion (BM)
Brownian motion is also known as diffusion. Applied in the phylogenetic context (Felsenstein 1973 S. Am. J. Hum. Genet. 25:471-492) it represents a “random walk” along the branches of the phylogeny. This is the simplest version of the model where the trait value is assumed to change at random, in both rate and direction, in a given time period. Across the tree, the mean change is zero, so the expected variance in trait values at the tips is proportional to the amount of shared phylogenetic history between species or the sum of the branch lengths. This can be intepreted as traits changing only through genetic drift and is also known as the neutral drift model. Traits values of closely related species should be more similar than distantly related species. It is usually the null model of trait evolution, against which observed trait values are compared to quantify phylogenetic signal. The basic Brownian motion model has been extended many times, by the addition of one or more parameters, usually to transform the branch lengths in some way that tries to account for a particular evoutionary process. Some of these extensions are described below. For those transformations, When the \(\lambda\), \(\delta\) or \(\kappa\) parameters equal 1, the models predict Brownian motion.
Ornstein-Uhlenbeck model (OU)
The Ornstein-Uhlenbeck model (Lande 1976 Evolution 30:314-334; Butler and King 2004 . Am. Nat. 164:683-695) is an extension of the Brownian Motion model with an additional parameter, \(\alpha\), that attempts to model the strength of stabilising selection, or the pull, towards the trait’s optimum adaptive value, \(\theta\). The strength of the stabilising selection is assumed to be shared across a clade or some related subset of species. In theory, if the OU model fits the data better than th BM model, this could be interpreted as evidence for stabilising selection on that trait. However, in practice, there are other evolutionary or environmental factors that could generate the patterns expected under an OU model, for example, phenotypic plasticity.
Early-burst model (EB)
The Early Burst model (Harmon et al. 2010 Evolution 64:2385-2396) is also known as the accelerating-decelerating model (Blomberg et al. 2003 Evolution 57:717-745) and assumes that trait evolution reflects rapid diversification (adaptive radiation) in the basal part of the tree, followed by reduced rates of evolution as increased species richness generates stronger interspecific competition. The model has a rate change parameter, a, defining the the rate of evolution through time. This rate is assumed to decrease or increase exponentially through time. The \(\delta\) transformation of the Brownian Motion model (see below) is related to the Early Burst model. It is assumed that some ecological opportunity to exploit new niches underpins rapid diversifications/radiations, for example invasion, extinction of competitors or a novel innovation (Ingram et al. 2012 J. Evol. Biol. 25: 1902-1910). For example, whole genome duplication events are hypothesised to underpin the diversification of Phytophthora as a genus (Seidl et al. 2012 Genome Biology and Evolution, 4, 199-211) and, within the genus, the adaptive radiation that generated P. infestans and and many of it’s close relatives. It is hypothesised that the expansion of gene families directly or indirectly involved in pathogenicity allowed Phytophthoras to exploit new hosts and diverge into distinct species.
Trend
The trend model (Hansen and Martins 1996 Evolution 50:1404-1417) extends the Brownian Motion model so that the mean rate of trait evolution (0 in the BM model) is assumed to show a linear trend through time (either increasing or decreasing). It has one additional parameter for the slope of this trend. The expected mean trait value at time t is normally distributed with mean proptional to time. A good fit of this model can be interpreted as evidence for constant directional selection.
lambda
Lambda () is an additional parameter to the Brownian Motion model which estimates the linear transformation of the branch-lengths of the phylogeny that best reproduce the observed trait values. The lambda parameter is often used as a measure of phylogenetic signal. When \(\lambda = 1\) the model predicts the same distribution of trait values as the Brownian Motion model. When \(\lambda = 0\) the trait value has no phylogenetic structure and is distributed randomly across the tree. However, there are multiple evolutionary and ecological processes that could disrupt trait covariance among closely related species produce a departure from \(\lambda = 1\). Although it can’t be used to infer the specific evolutionary processes at play, it can be used to distinguish traits which seem to be either phylogenetically conserved / undergoing gradual divergence via drift (strong phylogenetic signal) compared to those traits that appear to be more labile (weaker phylogenetic signal). \(\lambda\) has been shown to be a robust statistical descriptor of phylogenetic signal under a range of trait evolution models (M?nkem?ller et al. (2012). Methods in Ecology and Evolution, 3, 743-756) and even when trees contain polytomies and are not accurately time-calibrated (Molina-Venegas, R. & Rodr?guez, M. (2017). BMC Evolutionary Biology, 17, 1-10).
kappa
Kappa (Pagel 1999) is an alternative additional parameter to the Brownian Motion model, which raises all branch lengths to a power (non-linear). This assumes that trait divergence depends on the number of speciation events between species and can be used to test whether evolution is punctuational (short periods of rapid evolution followed by longer periods of stasis in which no change occurs) or gradual (trait divergence accumulates steadily through time). The value of kappa stretches or compresses individual phylogenetic branch lengths. If \(\kappa > 1\), then long branches are stretched more than shorter ones, indicating evolutionary rates accelerate in longer branches. If \(\kappa < 1\), longer branches are compressed more than shorter ones and evolunationary rates decelerate in longer branches. When \(\kappa = 0\), trait evolution is independent of branch lengths and observed patterns are consistent with punctuated equilibrium. When \(\kappa = 1\) the tree is untransformed and equivalent to the Brownian Motion model.
delta
Delta is also an additional parameter to the Brownian Motion model which transforms the node depths of the tree, by raising them to a power. The delta parameter can be interpreted as an estimate of the relative contribution of early versus late evolution to the covariance between observed trait values and is similar to the EB model in this respect. A \(\delta = 1\) indicates gradual evolution, as predicted under Brownian Motion. A \(\delta < 1\) indicates temporally early trait evolution or ‘early burst’, indicative of an adaptive radiation. A \(\delta > 1\) indicates trait evolution has accelerated more recently, indicative of species-specific adaptation. In theory, I guess we could use this parameter to explore whether there is evidence for very recent rapid increases in rates of trait evolution for any of the Phytophthora traits? For example is \(\delta\) especially high for traits that might facilitate transport to, establishment and spread in novel regions and the expansion of host range?
drift
The directional drift model is another variant of the Brownian Motion model, which assumes weak selection on the focal trait. Here, the stregth and direction of selection varies randomly from one generation to the next. Selection in each generation is drawn from a normal distribution with mean 0 and variance parameter estimated by the model.
white
This is a white noise model, which assumes that trait values vary independently around some global mean, so phylogenetic history is completely irrelevant to observed trait values. This equates to \(\lambda = 0\), where the phylogeny is star-shaped.
These are all off-the-shelf methods for fitting phylogenetic trees in a unified framework that means we can compare model performance. The cost of this is that we don’t have much flexibility in the types of models that we choose to compare. There are other models of trait evolution focussing on different evolutionary processes (e.g. biotic interactions, see Clarke et al. 2017 Am. Nat., 189: 121-137) that we haven’t tested.
Visualise how the individual traits are distributed across the phylogeny and quantify their phylogenetic signal
Fig. 3 - Distribution of trait values for 177 Phytophthora species grouped by A) sporangial features, B) temperature features, C) reproductive and survival features and D) oospore features. Grey regions indicate species for which trait data is unavailable. White regions indicate the focal structure is absent in that species. The best-supported models of trait evolution were based on linear (\(\lambda\)) transformations of branch lengths within the phylogeny and were used to infer the strength of phylogenetic signal in traits, \(\lambda\). When \(\lambda\) = 1, trait divergence is proportional to shared phylogenetic history (Brownian Motion). When \(\lambda\) = 0 the trait value has no phylogenetic structure and is distributed randomly across the tree. P values indicate whether the observed trait distribution was significantly different to that expected under Brownian motion (PBM) or a white noise model (Prandom).
Partitioning trait variation within and between clades
Relative disparity is a way of measuring trait disparity contained within sub-clades as a proportion of that contained across the entire clade. This can be mapped through time to explore the tempo of trait evolution. High levels of relative disparity indicate that sub-clades have independently evolved to have largely overlapping trait values across clades. If relative disparity is low relative to expectations under Brownian Motion then sub-clades have narrower and more distinct trait values.
For individual traits
Fig. 4 Phytophthora trait disparity through time (temporal patterns in trait diversification) for all traits grouped by A) sporangial features, B) temperature features, C) reproductive and survival structures and D) oospore features. Relative time describes the node height in phylogenetic history from root (0) to tip (1). Solid black lines are empirical trait disparity. Dashed black lines indicate the expected mean disparity under Brownian Motion (a null model). The grey shaded area represents 95% confidence intervals for 1000 simulations under the null model. Where solid black lines fall outside of the grey shaded region, disparity is significantly different from null model expectations. The morphological disparity index (MDI) is the area between the empirical and the mean expected disparity curves. MDI is a summary statistic for comparing the disparity through time curves for different traits with significance determined using a global envelope test. Higher MDI indicates more labile traits.
Fig. 5 The most labile Phytophthora sporangial feature (A), temperature feature (B), reproduction and survival structure (C) and oospore feature (D) based on trait disparity through time (temporal patterns in trait diversification) and phylogenetic signal (??) using an ITS-only phylogeny. Relative time describes the node height in phylogenetic history from root (0) to tip (1). Solid black lines are empirical trait disparity. Dashed black lines indicate the expected mean disparity under Brownian Motion (a null model). The grey shaded area represents 95% confidence intervals for 1000 simulations under the null model. Where solid black lines fall outside of the grey shaded region, disparity is significantly different from null model expectations. The morphological disparity index (MDI) is the area between the empirical and the mean expected disparity curves. MDI is a summary statistic for comparing the disparity through time curves for different traits with significance determined using a global envelope test. Higher MDI indicates more labile traits. The trait values for all 179 species (see below for abbreviated species names) mapped on to the ITS phylogeny showing the 11 recognised Phytophthora clades.See below for abbreviated species names.
Table 1 Abbreviated species names in Fig. 5| species name | abbreviation |
|---|---|
| P. ‘acacia’ | aca |
| P. acerina | ace |
| P. agathidicida | aga |
| P. alticola | alt |
| P. amaranthi | ama |
| P. amnicola | amn |
| P. aquimorbida | aqm |
| P. ‘aquatilis’ | aqu |
| P. arenaria | are |
| P. asiatica | asi |
| P. asparagi | asp |
| P. attenuata | att |
| P. austrocedri | aus |
| P. balyanboodja | bal |
| P. bilorbang | bil |
| P. bisheria | bis |
| P. boehmeriae | boe |
| P. boodjera | boo |
| P. borealis | bor |
| P. botryosa | bot |
| P. brassicae | bra |
| P. cactorum | cac |
| P. cajani | caj |
| P. cambivora | cam |
| P. capensis | cap |
| P. caryae | car |
| P. castaneae | cas |
| P. chlamydospora | chl |
| P. chrysanthemi | chr |
| P. cichorii | cic |
| P. cinnamomi | cin |
| P. citrophthora | cip |
| P. citricola | cit |
| P. clandestina | cla |
| P. constricta | cnt |
| P. cocois | coc |
| P. colocasiae | col |
| P. condilina | con |
| P. cooljarloo | coo |
| P. capsici | cpi |
| P. captiosa | cpt |
| P. crassamura | cra |
| P. cryptogea | cry |
| P. ‘castitis’ | csi |
| P. castanetorum | cst |
| P. dauci | dau |
| P. drechsleri | dre |
| P. elongata | elo |
| P. ‘emzansi’ | emz |
| P. erythroseptica | ery |
| P. estuarina | est |
| P. europaea | eur |
| P. fallax | fal |
| P. flexuosa | fle |
| P. fluvialis | flu |
| P. foliorum | fol |
| P. formosa | for |
| P. fragariae | fra |
| P. fragariaefolia | frf |
| P. frigida | fri |
| P. gallica | gal |
| P. gemini | gem |
| P. gibbosa | gib |
| P. glovera | glo |
| P. gonapodyides | gon |
| P. gregata | gre |
| P. gondwanensis | gwn |
| P. hedraiandra | hed |
| P. heveae | hev |
| P. hibernalis | hib |
| P. himalsilva | him |
| P. humicola | hum |
| P. hydropathica | hyd |
| P. hydrogena | hyg |
| P. idaei | ida |
| P. ilicis | ili |
| P. infestans | inf |
| P. insolita | ins |
| P. intricata | int |
| P. inundata | inu |
| P. ipomoeae | ipo |
| P. iranica | ira |
| P. irrigata | irr |
| P. intercalaris | itc |
| P. ‘kelmania’ | kel |
| P. kernoviae | ker |
| P. kwongonina | kwo |
| P. lactucae | lac |
| P. lateralis | lat |
| P. lacustris | lcs |
| P. lilii | lil |
| P. litoralis | lis |
| P. litchii | lit |
| P. macilentosa | mac |
| P. macrochlamydospora | mch |
| P. meadii | mea |
| P. medicaginis | med |
| P. megakarya | meg |
| P. ‘mekongensis’ | mek |
| P. melonis | mel |
| P. mengei | men |
| P. mexicana | mex |
| P. megasperma | mgs |
| P. mirabilis | mir |
| P. mississippiae | mis |
| P. morindae | mor |
| P. moyootj | moy |
| P. multivora | mul |
| P. multivesiculata | muv |
| P. nagaii | nag |
| P. nemorosa | nem |
| P. nicotianae | nic |
| P. niederhauserii | nie |
| P. obscura | obs |
| P. occultans | occ |
| P. ‘ohioensis’ | ohi |
| P. ornamentata | orn |
| P. pachypleura | pac |
| P. palmivora | pal |
| P. parvispora | par |
| P. ‘paludosa’ | pau |
| P. ‘personii’ | per |
| P. phaseoli | pha |
| P. pinifolia | pif |
| P. pini | pin |
| P. pisi | pis |
| P. pluvialis | pls |
| P. plurivora | plu |
| P. polonica | pol |
| P. porri | por |
| P. primulae | pri |
| P. ‘prodigiosa’ | pro |
| P. ‘parsley’ | prs |
| P. pseudocryptogea | psc |
| P. pseudolactucae | psl |
| P. parsiana | psn |
| P. pseudopolonica | psp |
| P. pseudorosacearum | psr |
| P. pseudosyringae | pss |
| P. pseudotsugae | pst |
| P. psychrophila | psy |
| P. pistaciae | ptc |
| P. quercina | quc |
| P. quercetorum | que |
| P. quininea | qui |
| P. ramorum | ram |
| P. ‘raspberry’ | ras |
| P. rhizophorae | rhi |
| P. richardiae | ric |
| P. riparia | rip |
| P. rosacearum | ros |
| P. rubi | rub |
| P. sansomeana | san |
| P. siskiyouensis | sis |
| P. sojae | soj |
| P. stricta | str |
| P. syringae | syr |
| P. tentaculata | ten |
| P. terminalis | ter |
| P. thermophila | the |
| P. trifolii | tri |
| P. tropicalis | tro |
| P. tubulina | tub |
| P. tyrrhenica | tyr |
| P. uliginosa | uli |
| P. uniformis | uni |
| P. versiformis | ver |
| P. vignae | vig |
| P. virginiana | vir |
| P. vulcanica | vul |
| P. ‘walnut’ | wal |
| P. x alni | xal |
| P. x andina | xan |
| P. x heterohybrida | xhe |
| P. x incrassata | xin |
| P. x multiformis | xmu |
| P. x pelgrandis | xpe |
| P. x serendipita | xse |
| P. x stagnum | xst |
Trait covariance across species using latent variable models
Latent variable models allow ordination of species in trait-space using a multivariate glm approach. This helps to cope with mixed trait variables (continuous, binary, discrete) as we can specify different error structures for different traits, preserve the mean-variance relationships for differently distributed traits. Compared to distance based approaches usually applied in ordination, this should improve the power to detect important axes of trait-space. We can also make use of model diagnostics based on residuals to ensure the fit of the model is acceptable and to quantify deviance explained. The estimated latent variables can be interpreted as unobserved variables and, in theory, as the ecological strategies underpinning observed patterns of trait covariance, like the axes of a PCA axes.
## Compiling model graph
## Resolving undeclared variables
## Allocating nodes
## Graph information:
## Observed stochastic nodes: 5955
## Unobserved stochastic nodes: 807
## Total graph size: 24534
##
## Initializing model
Fig. 6 Biplot of trait covariance among 179 Phytophthora species, with trait loadings (arrows) shown for the 20 most important traits. Phytophthora traits are coloured according to their hypothesised adaptive value for the three BAM niche processes (“filter”) in order to explore potential resource allocation trade-offs between investing in stress-tolerance traits to overcome environmental filtering (abiotic), combative traits to overcome niche partitioning (biotic) or dispersal traits (movement). Species (points) with more similar trait values are positioned closer together in trait-space. Species are coloured by whether they are predominantly known from soil, aerial or aquatic substrates to demonstrate how traits can mediate habitat filtering of dominant species. The axes can be interpreted in the same way as unconstrained ordination. The latent variables and the trait loadings (latent variable coefficients) were estimated using pure latent variable models with a species by trait matrix as the response.
How strong is the phylogenetic signal in the latent variables?
Latent variable 1: \(\lambda\) = 0.897
Latent variable 2: \(\lambda\) = 0.606
Mantel r = 0.244, p = 0.001
Table 2 Phytophthora trait lability and trait covariance. Phylogenetic signal, ??, scales from 0 to 1, where 1 indicates the distribution of trait values among species is entirely phylogenetically conserved. A value of 0 indicates trait values are distributed randomly and is evidence of extreme trait lability in response to selective pressures. MDI is the morphological disparity index developed by Harmon et al. (2003) which measures the difference between observed trait disparity and the disparity expected under Brownian motion (null model), using a global envelope test to determine if these differences were significant at any point in evolutionary history. MDI is a summary statistic for the temporal patterns in trait lability visualised in Fig. 3. Values of ?? and MDI significantly different from those expected under Brownian motion (null model) are denoted with * and are evidence of significantly labile traits. Trait covariance is quantified as shared responses to latent (unobserved) variables estimated using the latent variable coefficients (??LV1, ??LV2) from model-based ordination. Pink cells indicate positive responses and blue values indicate negative responses to latent variables. More similar colours indicate shared responses and evidence of covariance among those traits. Latent variable coefficients that do not overlap zero are denoted * and are interpreted as evidence for important trait loadings on that latent variable.| sporangial features | ||||||||
| caducous | 0.771 | * | -0.134 | -1.790 | * | 0.086 | ||
| papillate | 0.942 | 0.019 | NA | NA | ||||
| non-papillate | NA | NA | NA | NA | 4.652 | * | -1.406 | |
| papillate | NA | NA | NA | NA | -0.689 | * | 0.234 | |
| semi-papillate | NA | NA | NA | NA | -1.361 | * | 0.520 | * |
| proliferating | 1.000 | -0.262 | 5.306 | * | -1.755 | * | ||
| sporangiophore form | 0.650 | * | -0.189 | NA | NA | |||
| closed sympodium | NA | NA | NA | NA | -0.648 | -0.131 | ||
| compound sympodial | NA | NA | NA | NA | -1.477 | * | -0.031 | |
| downy mildew shape | NA | NA | NA | NA | 0.093 | 1.139 | ||
| irregularly branched | NA | NA | NA | NA | -0.375 | -0.018 | ||
| simple sympodial | NA | NA | NA | NA | -1.046 | * | 0.642 | * |
| unbranched | NA | NA | NA | NA | 1.906 | * | -0.758 | * |
| temperature features | ||||||||
| growth rate at optimum temp | 0.943 | * | 0.082 | 0.999 | * | -1.053 | * | |
| max temp for growth | 0.873 | * | -0.017 | * | 2.497 | * | -2.470 | * |
| min temp for growth | 0.635 | * | 0.250 | * | -0.057 | -0.384 | ||
| opt temp for growth | 0.873 | * | 0.024 | * | 1.985 | * | -2.161 | * |
| temp range | 0.610 | * | 0.219 | * | 2.502 | * | -2.117 | * |
| reproductive and survival features | ||||||||
| chlamydospores | 0.641 | * | -0.034 | -0.055 | -0.244 | * | ||
| reproductive strategy | 0.784 | * | -0.066 | NA | NA | |||
| heterothallic | NA | NA | NA | NA | -0.369 | -0.961 | * | |
| homothallic | NA | NA | NA | NA | 0.952 | 5.104 | * | |
| sterile | NA | NA | NA | NA | 0.491 | -4.733 | * | |
| hyphal swellings | 0.820 | * | -0.120 | 0.538 | * | -0.055 | ||
| oospore features | ||||||||
| antheridial attachment | 0.873 | * | -0.037 | NA | NA | |||
| amphigynous | NA | NA | NA | NA | -0.882 | * | -2.593 | * |
| amphi- and paragynous | NA | NA | NA | NA | 0.082 | 0.724 | * | |
| paragynous | NA | NA | NA | NA | 0.429 | 0.819 | * | |
| space between oospore and oogonial wall | 0.624 | * | -0.162 | NA | NA | |||
| aplerotic | NA | NA | NA | NA | 0.245 | * | 0.009 | |
| plerotic and aplerotic | NA | NA | NA | NA | -0.039 | 0.125 | ||
| plerotic | NA | NA | NA | NA | -0.233 | -0.128 | ||
| slightly aplerotic | NA | NA | NA | NA | -0.030 | 0.156 | ||
| oogonial wall | 0.658 | * | -0.024 | NA | NA | |||
| ornamented | NA | NA | NA | NA | 0.210 | -0.800 | * | |
| smooth | NA | NA | NA | NA | -0.512 | * | 0.812 | * |
| wavy | NA | NA | NA | NA | 2.186 | * | -0.717 | |
| oospore wall index | 0.435 | * | 0.385 | * | 0.043 | * | 0.011 | * |
| oospore wall thickness | 0.423 | * | 0.388 | * | 0.512 | * | 0.000 | |
| oospores | 0.896 | 0.215 | -0.473 | 4.780 | * | |||
| tapering base | 0.857 | -0.053 | 0.212 | 0.068 | ||||
The negative correlation (r = -0.536, p = 0.018) between phylogenetic signal (\(\lambda\)) and morphological disparity index (MDI) indicates consistency among both metrics of trait lability